Identifying Layers Susceptible to Adversarial Attacks